DEVELOPMENT OF PRESSURE ESTIMATOR AND VELOCITY FIELD CORRECTIONS FOR PARTICLE IMAGE VELOCIMETRY USING PHYSICS-INFORMED NEURAL NETWORK

• Main Author: Christian Chandra, Calvin
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/66857
• Institution: Institut Teknologi Bandung

Flow diagnostics using particle image velocimetry (PIV) has always been a viable option, but errors or faults in the experiment can lead to misinterpreted data. Meanwhile, physics-informed neural network (PINN) usage has been on the rise because of its versatility. This work intends to analyze the possibilities of implementing PINN for PIV and test the initial program on a couple of flow cases to observe whether misinterpretations in PIV output can be minimized, along with providing pressure prediction in the analysis domain. This work modifies an already existing PINN program to better suit PIV applications which is then implemented on backstep flow and flow behind a cylinder as the test cases. Initial results show that the PINN is capable of filling in gaps of missing data and correct invalidly measured data to a certain extent. The backstep flow test case reveal that the PINN can appropriately represent the velocity vectors, but not entirely if false input data is present. However, the pressure contours of this case are not entirely certain. Meanwhile, the PINN can satisfactorily represent the velocity vectors and pressure contours for flow behind a cylinder that does not have any data alterations, at least qualitatively. This brings forth the conclusion that this initial stage of the PINN program, while having some pleasing results, emphasize the need of further research in developing and generalizing the proposed PINN program.